Answer Box: Open-Inspect, an open-source background agents coding system inspired by Ramp's Inspect, has been released, enabling developers to offload coding tasks to an AI agent that works asynchronously in the background. The system supports multiple AI models, real-time multiplayer collaboration, and integration with Slack, GitHub, and Linear, but is designed for single-tenant deployment within trusted organizations.
What is Open-Inspect?
Open-Inspect is a hosted background coding agent that allows developers to assign tasks and continue working while the agent executes autonomously. It provides access to full development environments including Node.js, Python, git, browser automation, and VS Code. The system can be accessed from anywhere via a web UI, Slack, GitHub PRs, Linear issues, or webhooks, and supports real-time multiplayer sessions where multiple users can collaborate simultaneously.
Key Features and Architecture
The agent can create PRs with proper commit attribution to the prompting user, run on a schedule via cron jobs, Sentry alerts, or webhook-triggered automations, and spawn parallel sub-tasks that work in separate sandboxes concurrently. Users can choose their preferred AI model, including Anthropic Claude, OpenAI Codex (via ChatGPT subscription), or OpenCode Zen.
The architecture consists of a control plane hosted on Cloudflare using Durable Objects for session management, WebSocket connections, and event streaming, along with a D1 database for repo-scoped secrets. The data plane runs session sandboxes that include a supervisor, OpenCode, and a bridge connecting back to the control plane. Sessions start near-instantly through filesystem snapshots, pre-built images rebuilt every 30 minutes, and proactive warming that begins spinning up the sandbox as soon as the user starts typing.
Single-Tenant Design and Security Considerations
Open-Inspect is designed for single-tenant deployment only, where all users are trusted members of the same organization with access to the same repositories. It uses a shared GitHub App installation for git operations, meaning all users share the same GitHub App credentials. The system does not perform per-user repository access validation before creating a session. For PR creation, GitHub users' OAuth tokens are used for proper attribution, while users signing in via other methods fall back to the shared GitHub App bot.
For organizations concerned about security, the documentation recommends deploying behind SSO/VPN, installing the GitHub App only on intended repositories by selecting specific repositories rather than "All repositories," and restricting sign-in by configuring allowed GitHub users, email domains, or active GitHub organization membership. For multi-tenant deployment, the system would require per-tenant GitHub App installations, access validation at session creation, and tenant isolation in the data model.
Integration and Automation
The system integrates with multiple platforms: a web UI with real-time streaming, terminal panel, and multiplayer presence; a Slack bot that can be @mentioned or DM'd to start a session; a GitHub bot for auto-review on PR open or responding to @mentions in PR comments; a Linear bot that can be mentioned or assigned on an issue to start a coding session; and webhooks for triggering sessions from any external system via authenticated HTTP POST.
Automation features include cron schedules with timezone support, Sentry alerts for auto-triage on new errors or regressions, and inbound webhooks. Sessions can work across up to 10 repositories simultaneously in a single sandbox, with the ability to make coordinated changes and open a PR per repository. Environments allow saving a repository set as a named environment with its own secrets scope and optional prebuilt images.